Pressure Drop Analysis of Direct Liquid Cooled (DLC) Rack (ITherm 2017)

Overview

This record consists only of the metadata (abstract) of a paper presented at IEEE ITherm 2017 (DOI 10.1109/ITHERM.2017.7992570) by Sami Alkharabsheh, Bharath Ramakrishnan and Bahgat Sammakia; the full text has not been read. The paper measures and models flow and pressure distribution in a commercial direct-liquid-cooled rack comprising a coolant distribution module, supply/return manifolds and server cold-plate modules. The authors built pressure-flow curves for each component and an experimentally validated flow network model.

The model errors are 2.5% for server-module flow and 3.8% for coolant distribution module flow. This shows that the validated flow network model predicts flows within a few percent. According to the results, server modules account for 56% of total rack pressure drop. Of the server-module pressure drop, 69% comes from plumbing (corrugated hoses, quick disconnects, fittings) rather than the cold plate.

The measured flow maldistribution among servers was attributed to variation in micro-channel cold-plate manufacturing. The study was carried out on a single commercial rack architecture, so the percentages are specific to that rack and cannot be generalised to every system. Because the record relies only on the abstract, details of the method and operating conditions are not given here.

Key points

1. Flow network model errors: 2.5% for server-module flow and 3.8% for coolant distribution module flow.
2. Server modules account for 56% of total rack pressure drop.
3. 69% of server-module pressure drop comes from plumbing (corrugated hoses, quick disconnects, fittings) rather than the cold plate.
4. Measured flow maldistribution among servers was attributed to micro-channel cold-plate manufacturing variation.
5. Pressure-flow curves were built for each component, with an experimentally validated flow network model.
6. The study rests on a single commercial rack; values cannot be generalised.
7. The record rests only on the abstract; the full text was not read.

Using it in design

• Include hose, quick-disconnect and fitting losses explicitly in the tool's rack-level hydraulic calculations, not just cold-plate pressure drop.
• Treat a flow network solver as accurate enough for design-stage flow balancing.
• Allow for server-to-server flow maldistribution in the design margin.
• Assess quick disconnects and hose selection for their effect on total pressure drop.
• Verify the values for your own rack architecture against manufacturer data.

Who should read it

Mechanical engineers and thermal analysts designing direct-liquid-cooled rack and manifold hydraulics.

Publication details

Authors

Sami Alkharabsheh; Bharath Ramakrishnan; Bahgat Sammakia

Publisher

IEEE

Year

2017

Document number

DOI 10.1109/ITHERM.2017.7992570

Event

IEEE ITherm 2017

Document language

English

Source type

Standard

Access

Citation only

Licence

Copyright retained; citation only — the publisher’s terms apply to the original document

Citation